Research On Object Identification For Soccer Robot

With the rapid development of artificial intelligence, computer vision is playing an increasingly important role in daily life. For the soccer robots, in order to cooperate and compete like humans, its visual information processing is the key technology. In this thesis, taking the humanoid league of RoboCup competition as the study object, which is the mainstream robot soccer competition nowadays, our research target is recognition algorithms of soccer robot.In the environment of RoboCup humanoid league, there exist some external interferences in vision system, such as light impact, image quality, etc. Identifying team members of both sides is one of the difficulties as well. While the competition’s pleasing of eyes is asking for highly real-time visual information processing algorithms. Therefore, this thesis focuses on accuracy and efficiency of feature extraction algorithms. The main work is as follows:In order to improve the focalization of target identification, the Mean Shift-SIFT method was presented in Chapter3, which preprocesses image at first, then extracts features. Specifically, we use Mean Shift algorithm to extract the main part firstly, then use the SIFT algorithm to extract the feature points, and match feature points using the nearest neighbor method finally. Experimental results illustrated that the proposed method improved the matching degree comparing with the traditional method at different specific interferences, and increased the processing speed.As a feature extraction algorithm, the SIFT algorithm can identify team members from the opposite ones, but the algorithm itself is time-consuming, and will directly extend the matching time. An improved object recognition algorithm (called PCA-SIFT algorithm) was proposed in Chapter4, in which principal component analysis method (PCA) is applied on the128-dimensional descriptors of the classical SIFT algorithm to reduce the dimensionality, and the nearest neighbor method is used to match feature points. Experimental results illustrated that the PCA-improved descriptors improved computational efficiency while remaining high matching degree.By simulating the interference scenario of soccer robot competition they may encounter, we confirmed the effectiveness of target feature extraction of the Mean Shift-SIFT method and PCA-SIFT method. These target feature extraction methods made the foundation of decision-making and motion control in robot soccer competition.